Master's thesis: Defect detection for foundries using advanced computer vision systems

Master's thesis: Defect detection for foundries using advanced computer vision systems

Introduction:
Casting defects are an inevitable part of the casting process where a liquid material is poured into a mould that contains a hollow cavity of the desired shape. There are many kinds of casting defects, including cracks, blow holes, pinholes, burrs, shrinkage flaws, flaws in the mold material, pouring flaws, metallurgical flaws, etc. One the key issues has been the time-consuming inspection process which is often carried out manually often leading incorrect rejection due to human accuracy. By utilizing automated quality inspection systems, companies can reduce the time taken to identify and reject defective products, maximizing thus the accuracy of the process.

Computer vision has grown rapidly in recent years, driven by advances in AI. In this thesis project, the student will develop, train, and evaluate an AI-based solution for detecting surface defects on cast metal components, using an industrial-grade camera and open-source models at Combi Wear Parts in Ljungby. The work includes image annotation, model selection, and on-site training. Several approaches will be tested, as different model types and data preparation strategies may offer varying strengths for defect detection.

Research goals:
The goal of this thesis project is to develop and optimize an AI-based computer vision solution for detecting surface defects on cast metal components. Instead of refining an existing commercial system, the project will focus on training and customizing open-source models to achieve high accuracy using data collected with an industrial-grade camera. Tasks include:

• Identification of key defects and dataset collection
• Investigating the need for and potentially generating synthetic datasets
• Training and evaluating suitable AI CV models, such as object detection and image segmentation models
• Deploying the models and evaluating their performance in real production

Methodology:
The project will focus on training and customizing open-source models to achieve high accuracy using data collected with an industrial-grade camera. Various machine learning (ML) techniques will be evaluated to enhance model performance. The models will be trained on different types of data, with particular focus on features such as surface defects, cracks, and discolorations.

General information about the project:
Start and end date: The thesis will be carried out during the spring semester of 2026, but the exact start and end dates will be determined by the relevant university and whether the level is bachelor or master.

Degree level:
The project is preferably carried out at master's level or similar but can also be suitable at bachelor's level.

Location:
Partly flexible.
Practical parts of the work must be done at the foundry in Ljungby.
Supervisor
(chosen by the university in question): XXX, YYY University, email

Co-supervisors:
Anton Ekman, Combi Wear Parts,

Andreas Thore, RISE,

Lennart Elmquist, RISE,

Examiner:
(chosen by the university in question): XXX, YYY University, email

Compensation:
1330 SEK/hp/student if one student, 1000 SEK/hp/student if two.

About Combi Wear Parts
Combi Wear Parts is a global export company that develops and manufactures its own patented wear part systems for the construction, mining, and dredging industries. Through its subsidiary, Combi Casting, the company also develops and produces critical key components on behalf of world-leading OEMs in the forestry and material handling industries.

The Combi Wear Parts Group headquarters is located in Kristinehamn, with production facilities in Ljungby and sales offices in the USA and Germany. The group employs approximately 130 people and has an annual turnover of about SEK 250 million. Combi Wear Parts is the largest independent steel foundry group in the Nordic region.

Learn more at

Welcome with your application